Most companies still argue about remote work as a question of location: home or office, two days or five. The firms that have run distributed at scale for years treat that framing as a category error. The variable that decides whether distributed work functions is not where people sit. It is whether the organization runs on documentation and judges output rather than presence.
Presence is a weak proxy, and the evidence knows it
The strongest field evidence on remote work is not about location preference. It is about what happens when you stop measuring attendance. In a randomized trial at the Chinese travel company Ctrip, call-center staff assigned to work from home became about 13 percent more productive, and attrition fell sharply (Bloom, Liang, Roberts and Ying, 2015). A later randomized trial at the same firm, published in Nature, tested the hybrid arrangement most companies now fight over: two days at home cut voluntary attrition by roughly a third, from 7.2 to 4.8 percent, with no measurable loss to performance or promotion over two years (Bloom, Han and Liang, 2024). The productivity and retention did not come from the building. They came from managing what people produced instead of when they were visible.
That shift, from presence to output, is the actual content of distributed work, and it is a management decision, not a real-estate one. Once you concede that good work can be defined and measured by results, the location question mostly answers itself.
The written record is the operating system
What replaces the hallway conversation and the over-the-shoulder question is not a chat app. It is documentation. GitLab, which runs one of the largest all-remote workforces, states the mechanism bluntly: it operates handbook-first, with communication documented in a public handbook that would run past two thousand printed pages, and it argues that asynchronous work is impossible without strong documentation, because without it a single follow-up question can block a colleague for hours (GitLab handbook, 2026). The handbook is not bureaucracy. It is the shared memory that lets people act without waiting for each other, which is the entire promise of async.
This reframes the value of writing things down. In a co-located office, undocumented knowledge is an inconvenience. In a distributed one, it is a hard dependency: every fact that lives only in one person's head becomes a bottleneck the moment that person is asleep, offline, or in another time zone. Documentation is the infrastructure that converts a scattered group into a coordinated one.
Autonomy comes in levels, and async is near the top
Matt Mullenweg, who runs the distributed company Automattic, offers a useful ladder for this. In his five levels of autonomy, the fourth level is defined by asynchronous work evaluated by output rather than by hours or process, and he argues that an organization operating at the highest level consistently outperforms office-bound rivals (Mullenweg, 2020). This is a practitioner's framework, not a controlled study, and it should be read as such. But it names the destination precisely: the goal is not remote as a benefit, it is asynchronous work judged by results, with real-time coordination reserved for the few things that genuinely need it.
The trap of remote without the redesign
The companies that tried remote and hated it usually kept every office habit and moved it onto video. They measured presence through green status dots, ran the same synchronous meeting schedule across time zones, and left knowledge undocumented. That is not distributed work. It is surveillance with a worse commute, and it fails for reasons that have nothing to do with whether distributed work can succeed. The Bloom trials show it can. GitLab and Automattic show how: change what you measure and write everything down.
The practical read
Stop debating desks and start redesigning two things. First, measurement: define and judge output, which the randomized evidence shows preserves performance while cutting attrition (Bloom, Han and Liang, 2024). Second, documentation: make the written record the default, because async coordination is impossible without it (GitLab handbook, 2026). Treat async as a level of organizational maturity to climb toward, not a perk to ration (Mullenweg, 2020). Remote work is not a place you let people go. It is a discipline of writing things down and trusting results, and companies that never build that discipline will keep concluding, wrongly, that distributed work does not work.
Sources
- Bloom, Liang, Roberts and Ying (2015). Does Working from Home Work? A randomized trial where home work raised productivity about 13 percent and cut attrition. https://academic.oup.com/qje/article-abstract/130/1/165/2337855
- Bloom, Han and Liang (2024). Hybrid working from home improves retention without damaging performance. A randomized trial where two days at home cut attrition by a third with no performance loss. https://www.nature.com/articles/s41586-024-07500-2
- GitLab handbook (2026). Handbook-first and asynchronous work. Documents that async work is impossible without strong documentation, positioning the handbook as the operating foundation. https://handbook.gitlab.com/handbook/company/culture/all-remote/asynchronous/
- Mullenweg (2020). Five Levels of Autonomy. A practitioner framework defining level four as async work judged by output rather than hours. https://ma.tt/2020/04/five-levels-of-autonomy/